JPS5945586A - Character recognizing device - Google Patents

Abstract

PURPOSE:To raise a processing speed of character recognition, by selecting special character without comparing it with general character and recognizing special character independently. CONSTITUTION:On a basis of the unknown pattern of an unknown character 2 in the first RAM 5, it is discriminated whether the character is a general character or a special character. In case of a general character, the address of a general character processing routine is designated, and the feature sequence of the character is extracted on a basis of preliminarily set general character processing procedure, and a dictionary guidance processing, a dictionary collation processing, etc. are performed in accordance with a set search method, thus recognizing the character 2. In case of a special character, the address of a special character routine corresponding to an encoded signal is selected, and the feature sequence of the character is extracted on a basis of the prescribed special character processing procedure, and the dictionary guidance processing, the dictionary collation processing, etc. are performed in accordance with a set search method, thus recognizing the character 2.

Description

DETAILED DESCRIPTION OF THE INVENTION (A) Field of the Invention The present invention relates to a character recognition system ((1) for automatically reading and recognizing characters such as kana, alphabets, and numbers.

BACKGROUND OF THE INVENTION In recent years, character recognition devices have been actively developed as input devices for computers.Most of these character recognition devices perform all character recognition in the order outlined below.

First, a character recorded on a sheet of paper is subjected to optical weight conversion into one character 7-, J:, and further converted into a two-legged pattern of rljJOJ. Next, in order to effectively perform post-processing on this dichotomous pattern, 1JiJ processing is performed. That is, a series of processes including noise processing such as black dots on the paper and smoothing of single character/graphic boundaries are performed. Next, the recognition steps necessary to perform two-character recognition] Characteristics of the unknown character of 茌 (intersection 1 branch point, /L/-
A feature extraction process is performed to extract the number of strokes (length information of one stroke, etc.). Based on the extraction results, candidates are narrowed down to characters that have common characteristics among a large number of stored characters.

After this narrowing down, the first stage recognition is completed, and if there are a plurality of candidate characters, a detailed identification process is performed to select the only character from among them.

This detailed identification process is called dictionary matching process, and this dictionary stores the characteristics of each character in a fixed series, and is usually composed of 9 ROMs. This narrowing down is called dictionary guidance processing, in which a specific character group is selected from the dictionary based on the feature group obtained by feature extraction.

Following the dictionary guiding process in this manner, a dictionary matching process is performed to detect the degree of coincidence between the standard feature series of each character (hereinafter referred to as a "marker pattern") stored in advance in the dictionary and the feature series of the unknown character.

Based on the matching results, the read character can be specified, or the recognition result can be output as unrecognizable.

In this character recognition device, the conventional + +iiJ unknown characters are treated with the same dictionary guidance processing and dictionary matching processing for each character without exception. -r deshi.

1. Periods, commas, and other special characters (including special characters like “a” and “r”)
It is processed through the same processing procedure as general characters such as AJ. Therefore, even for special characters having characteristics in terms of pattern, each character group is selected after a large number of feature extraction processes similar to those for general characters are performed on each character 111.

In this case, all parameter checks that do not originally require feature checking are performed for special characters, which wastes time for processing fj:, processing "1", which takes a long time to fill in the recognition queue. ” There was a problem that it hindered star formation.

(c) Purpose of the Invention In view of the above, the present invention provides a literary magazine gf'fk device which aims to shorten recognition processing time and speed up processing by distinguishing and recognizing special characters and general characters. This is what we provide.

B) Structure and effect of the invention In order to achieve the above object, the present invention provides a reading means for reading an unknown character pattern, and a reading means for reading an unknown character pattern, and a reading means for reading a circumscribed quadrilateral formed through the outermost point of the unknown pattern of the read unknown character. range detection means for detecting the size by comparing it with a preset size; position detection means for detecting the position of the read unknown pattern relative to the reading target area of the reading means; A separation detection means for detecting separation of characters, a length detection means for detecting the length of the circumscribed quadrilateral 'ff: @, and a recognition processing procedure for only special characters such as periods, commas, and bars are memorized, and special characters are stored. Memorize means for performing recognition processing and general character recognition processing procedures;
Depending on one combination of the means for performing general character recognition processing and the detection outputs of the range detection means 9 position detection means, distance detection means, and length detection means, the special character recognition processing means specified in r2TI or the general character recognition specified in MiJ is performed. Place 3! l! and means for selectively operating the execution means, and is configured to recognize special characters separately from general characters.

Therefore, according to the present invention, the special characters 1--are selected without comparing with the crotch characters, and! 1. Since the special characters are recognized separately, the processing speed of nine character recognition can be increased.

(E) Description of the Embodiment FIG. 1 shows a circuit block diagram of the character recognition device 1'1 of the present invention, in which the character 2 recorded on the paper P is read as an optical pattern by the 9' conversion circuit. Rare.

The photoelectric conversion circuit 4 connected to the photoelectric conversion circuit 11 converts the digital data back to 1 (1; 7aJ
t father means).

The A/D conversion circuit 4 is connected to the first RAM 5 (image memory), and together with the optical conversion circuit 15, the CPU 6
is connected to. This CPU 6 has a first RAM 5, a second RAM 7 (work memory), and ROMs 8, 9, 10, 1.
1.12, respectively.

As shown in FIG. The character pattern of the unknown character 2 to be recognized is stored two-dimensionally.

The leftmost X coordinate Xs, the rightmost X coordinate XE, the topmost Y coordinate Ys, and the bottommost Y coordinate YE of the character pattern 2a of the unknown character stored in the first RAM 5. are extracted and stored in the second RA'M7, respectively.9. These coordinates Xs
, XE, Ys, and Y+g, as shown in FIG. 6, a circumscribed quadrilateral 14 (sizes Xw, YII) passing through the outermost point of the unknown character pattern 2a is calculated.

In addition, the first ROM 8 stores the operating program of the CPU 6, and the jp, %2 ROM 9 has a period "・",
Comma ``,'', voiced mark ``'', half-voiced mark ``0''.

Criteria for detecting special characters such as bar “1” X1゜X3
．． Yl, Y2. Y3 is stored. This standard (l
fX "1. Y 1 is the X-direction value of the reference range (position C does not matter) 15 within the reading target area 13 set to detect the size of the unknown pattern 2a, as shown in FIG. is the Y direction value.

This standard fmX1. Y1 and Xw of circumscribed quadrilateral 14.

This will lead to exchange between China and YH. ,j,j,(Ill
2+'zY2 is the reading target area 1 as shown in FIG.
This is a value set to detect the entire position of the unknown pattern 2a relative to 3, and the entire reading target area 13 is divided into upper and lower halves.

It is detected whether the circumscribed quadrilateral 14 extends over the upper part, the upper part, or -1 octave. Fortune telling (?i, !4 value X mouth.

¥3 is the length of the outer j' quadrilateral 14 in the X direction, as shown in Fig. 6, and the length of the square, ii'il! This base? -m[DirectX3. This means that Y lo and Xw and YII of the circumscribed quadrilateral 14 are compared.

Mameko + ff1iJi 1st ROM 8 contains, along with other operation programs, a two-dimensional scanning operation program that scans the complete moon or target area 13 in the X direction and moves the scanning vertical direction in the Y direction, as shown in Fig. 7. By this scanning, the separation between the upper and lower strokes of the unknown character 2 is detected. In this case, Fig. 8(a), (
1)), Considering the Bunko pattern example shown in (0), the 8th
Figure (a) has no separate part, Figure 8 (b) i has a separate part, and Figure 8 (c) has two parts, but the left end of the upper part and the right end of the lower part overlap in the X direction. Since they fit together, there are no separates.

Ig3ROM10 is 1 classification table memory.

As shown in the right column of FIG. 9, error codes and addresses of various processing routines are written down. And, as shown in the left column of Figure 9, the size of the χ quadrilateral.

A coded signal that is sized according to the length position and the presence or absence of separation is added to a second RAM (work memory) 7. An address corresponding to the coded signal is specified, and an error code, reject code, or processing routine address is written in the specified address.
Renu is now output. For example, 2nd RA
If the coded signal from M7 is 00101100, address 44 will be specified and the dictionary processing routine address tt+1'' will be output.

The second RAM 7, as shown in FIGS. 10 and 11,
XS of the circumscribed quadrilateral 14 from which the unknown character pattern was extracted,
XE, YS, YE, size 7) r-code conversion results, etc. are stored, and it is a valley word record with addresses from 0 to 7. 1 word has 32
It has come to appear in bits. At address 0, outside jγ, Xs of quadrilateral 14.

Address 1 is XIL Atreson is Ys, A1 Runu 3
, address 4Kll-1, X)) and address 4Kll-1,

X w = X + Lu (XE-1) Y u = Y 1< -(Y s -
1) Furthermore, Atrenu 6 has separate %J, 1jj
The X-direction OR 1-direction for adding 1'11 to F is stored, and at address 7, the bits 1-24 to 1-2 are
6. Upper, middle, and lower digits of circumscribed quadrilateral 14: 1 piece Pi 1
-27 indicates presence or absence of separate, bit 28 to bit 1
91・Size and length of tangent quadrilateral 14 (Y11≦Y3
．． Y.

≦Y1. Xw≦X3+Xw≦X1) is Sera 1-
It is now possible to do so.

In addition, the 4th ROM11 is a sentence of general characters! The recognition processing routines such as character matching of special characters are stored in the 5th ROM 12, and the general character recognition processing routines and special character recognition processing routines stored in these Rotll. The selection operation is performed in response to an address signal from the ROM 10.

Next, a single character recognition operation will be explained.

First, the CPU 6 operates according to the program in the first ROM 8, and under the control of this CPU 6, the character 2 on the paper P is converted into an electrical signal by the photoelectric conversion circuit 6, and then the A/D
The conversion circuit 4 converts the decimal/I/f= number, and the black background of the reading target area 16 becomes 1. A white background is represented by O, and one character pattern is stored in the first RAM 5 (see FIG. 2).

Unknown pattern 2 of unknown character 2 in this first RAM 5
Based on a, a general character or a special character is identified. This identification operation will be explained according to the control flow shown in FIG. Note that the reference size X1.x stored in the second ROM9. X5. y
i, Y2. It is assumed that Y3 is set as follows.

X1=9. X3=7°Yi=9. Y2=16. Y3=7° First, at step 211/C, '! ;IG+ RAIVI5
The locus 1 of the circumscribed quadrilateral 14 passing through the most Ku point in the XY direction in the unknown pattern 2a is measured J1. That is, the first
Change the Y values of the seats and gaps in RAM5 in ascending order (D, 0
).

(0,1), (0,2),...(0,31)
, (1,0), (1,1),...),
Black and white (1/10) all discriminated, and the first black (1) Pi) -'s + (51, RoO),
(31,29), ...... (Ro1.
0), (30,31), (ro0.30).

......), and determines black and white (1/)i, and first it is black (1) and at 7, the X coordinate of the bit 'fcX I・
: To be.

Furthermore, the values of the coordinates are sorted in order (0, 0
).

(1,0) , (2,0) ... ... (ro1
, O), (0,1), (1,1), ・mountain), and then in descending order of ox (31,3
1).

(30,51), (29,31), ...
...(0,31), (31,30n.

(3030), ......), and the Y coordinates of the first black (1) bit are Ys and YI, respectively.
>. These Xs, X1 bow, YS, and Yl (are the coordinates of the circumscribed quadrilateral 14, and are stored in addresses D to D of the second RAM 7, respectively.

Subsequently, in step 22, the above Xs and X1g.

Ys, Yx+Yoshi circumscribed quadrilateral 14 (unknown pattern 2a)
The width Xw and height YH of are calculated and stored at address 4.5 of the second RAM 7.

Subsequently, in step 23, the size and length of the circumscribed quadrilateral 14 are compared, i.e., Xw and Y IT
xl with a reference range of 15°16 stored in the second ROM 9
, yl and )l are compared to see if they are larger than Y5, and Y
H≦Y6. If YI (≦Y1.

Next, in step 24, the position where the unknown pattern 2a is read in the A quadrant area is detected, that is, the second
Ys and YE stored at address 2.3 of RAM 7 and Y 2 stored in second ROM 9,...! : is compared,
When Y E < (32-Y2), unknown pattern 2
Since a is located in the upper half, bit 24 of address 7 of the second RAM 7 is set to 1.

When Ys≧(32-Y2), the unknown pattern 2a is located in the lower half, and 1 is set in Rihino 1 to 26, and in other cases, the unknown pattern 2a is located in the lower half, and in other cases, the last 9':11 pattern 211 or the upper and lower parts are set. Therefore, 1 is set in hit 25.

Then, in the stem drive 25, the unknown pattern 2a's sepantate (is omitted). That is, in the reading target area 15 of the first RAM 5, the logic 'All' of the X coordinate is performed for each of the Y coordinates (0 to 61).

As a result, it is stored in each address 6 of the f-th RAM 7, and each bit whistle is checked to see that if they are 1-0-1, they are separated vertically.

i21 (bit 27 of address 7 of AM7 is set to 1t, and the characteristics of unknown character 2 are extracted.

Finally, in step 26 + 21st Z A+VI
7, coded at address 7 / coded 11〆1
Based on this, the error code, address code, and addresses of various processing routines stored in the third goMto are specified, and the processing operations of the corresponding processing routines are executed. That is, in the case of an error code, the coded value is logically 1 to not present, and in the case of a reject code, it is not to be read.

In the case of a special character, the address of the special character processing routine corresponding to the encoded number (for example, address 44.83 shown in FIG. 9) is selected and preset with this address. 15 a.

A feature sequence of one character is extracted based on the special character processing procedure of 12 characters per side, and dictionary guidance processing is performed according to the set search method.
Dictionary checking processing and the like are performed and one character 2 is recognized.

On the other hand, for general characters, the address of the general character processing routine (for example, address 64 shown in FIG. 9) is specified, and the 4H! The sequence is extracted, and according to the set search method, dictionary guidance processing, dictionary matching processing, etc. are performed to recognize eight characters.

Next, the discrimination operation between general characters and special characters will be explained for specific recognition target character 2 based on FIGS. 13 to 15.
5, a circumscribed quadrilateral 14 is drawn from the unknown pattern 2.t, and its height, U, and width are determined. The calculated value for each person is X5=5. X+Shiny 3 0 +Ys =20.
Yll; =22.

Xw=26.　　　　Yo=3　. It becomes.

These are xi =9. X'3=7. Y 1-9. Y2=
16. When compared with ¥3-7 and performing the X-direction logical sum of the address 6 of the second RAM 7, the coded value of the address 7 of the second RAM 7 becomes 44, and the
-"Jiwadokoro 1shi11 routine is selected.

In FIG. 14, a separate is detected, and the other circumscribed quadrilaterals 14 are measured in the same way, and the encoded value of address 7 of the second RAM 7 becomes 335, and the coded value of the third ROM 10 becomes 335.
The ";" dictionary no. 1 processing routine is selected.

In FIG. 15, address 7 of 2nd RA, IV'I7
The encoded value becomes 60, and the third ROM 400 general character processing routine is selected.

Note that the range detecting means for detecting the size of the unknown pattern 2a may calculate and compare one area in addition to comparing the lengths of Xl and Y'l as in the embodiment.

Still, the length detecting means of the unknown pattern 2a is X3. 'Outer 1 width (Xw
) and the height (YH) may be calculated and compared with the reference value.

Furthermore, it is not necessary to set only ¥2 as the reference value for the position changer of the unknown pattern 2a as in the embodiment.

Many types of reference values including the X direction may be set.

[Brief explanation of drawings]

The drawings show an embodiment of the present invention; FIG. 1 is a circuit block diagram of a character recognition device, FIG. 2 is an explanatory diagram of an unknown pattern stored in the first RAM, and FIG. 6 is an unknown pattern showing a circumscribed quadrilateral. 4 is an explanatory diagram of the reference value X1°Yl for comparing the size of unknown patterns, FIG. 5 is an explanatory diagram of reference value Y2 for detecting the position of unknown patterns, and Reference value for length comparison X3. Explanatory diagram of ¥3. FIG. 7 is an explanatory diagram of the running operation for detecting the separator 1-. Figures 8(a), (b), and (0) show the sudden appearance 1 of Separate I-. ). 9 is an explanatory diagram of 1 loquat and 1. stored in the 3rd ROIA. is an explanatory diagram of the contents, Figure 10 is the 2nd RA
114 is an explanatory diagram of the memory contents stored in the second
12 is a control flow diagram explaining part of the operation of the control 1II41 of the character recognition device, and FIGS.
The figure shows the identification mode for 44 characters, and Figure 16 (
a,), Fig. 14 (a,) and Fig. 15 (al is the lth
Explanatory diagram of unknown pattern of RAM, Fig. 13(b), 1st
FIG. 4 (1)) and FIG. 15 (1)) are explanatory diagrams of the storage contents of the second RAIVI. 1: Character recognition device, 2: Characters, 3: Photoelectric conversion circuit,
4: A/D father replacement b! ri+5・7: RAM,
6: CPU, (3, 9, 10, 11, 12: R
Old, 13: r' door collection elephant 1 reconnaissance area. 14: Circumscribed quadrilateral, 15・16: -1A quasi 0 (12
Enclosed. 2a: Unknown pattern. Patent Applicant Tateishi Seiki Co., Ltd. Agent Patent Attorney Shigeru Nakamura ID Figure 1 1 Figure 2 X Figure 3 Figure 6
N Fig. 14-α Fig. 15 (α) X→- Cb) Bit (b) Bit

Claims (1)

[Claims] (1) A reading means for reading an unknown character pattern, and a range detection means for comparing and detecting the size of a circumscribed quadrilateral formed through the outermost point of the unknown pattern of the read unknown character with a preset size. . position detection means for detecting the position of the read unknown pattern relative to the reading target area of the reading means; separation detection means for detecting separation of unknown characters by scanning the unknown pattern in one direction; and length detection means for detecting the length of the circumscribed quadrilateral. 9. Means for storing recognition processing procedures for only special characters such as periods, commas, and bars, and performing special character recognition processing, and storing processing procedures for general character recognition. Depending on the combination of the means for performing general character recognition processing and the detection outputs of the range detection means 1 position detection means, distance detection means and length detection means, the special character recognition processing means described in I8U or the general character recognition processing is performed. Choosing the means 1'
1. A character recognition device C1° characterized in that the character recognition device C1 is equipped with a lower stage that recognizes special characters separately from general characters.